Thief

The pairing of an Intel Core i3-12100F with an NVIDIA GeForce RTX 5090 in Thief presents a striking case study in system bottlenecking. The measured frame rates reveal a configuration where the GPU’s immense raw power is frequently held in check by the processor’s four-core design, yet the results remain remarkably playable across all tested configurations. This analysis examines how the data reflects the interplay between a 2014 stealth title and modern high-end hardware, focusing on where the performance ceiling actually lies.

Resolution Scaling

The frame rate progression from 1080p to 4K in Thief shows a clear pattern of diminishing returns, but with a twist that points directly at the CPU as the primary limiter at lower resolutions. At 1920x1080 with High settings, the system produces 138 FPS average. Moving to 2560x1440 at the same settings drops this to 122 FPS, a reduction of roughly 11.6%. The jump to 3840x2160 further lowers the average to 86 FPS, which represents a 29.5% drop from the 1440p figure. This scaling curve is unusual for a GPU as powerful as the RTX 5090, which typically shows more aggressive performance cliffs at 4K.

The telltale sign of CPU limitation appears when comparing Low settings across resolutions. At 1080p Low, the system achieves 169 FPS. At 1440p Low, this only falls to 149 FPS, and at 4K Low, it still manages 125 FPS. The delta between 1080p and 1440p on Low is just 11.8%, while the gap between 1440p and 4K is 16.1%. For a GPU with a 1.79 TB/s memory bandwidth and 32 GB of GDDR7 VRAM, this relatively shallow decline suggests that the processor is throttling the frame rate ceiling before the GPU can fully stretch its legs. If the RTX 5090 were the sole bottleneck, the 4K Low result would likely be far closer to the 1440p number, given the game's age and modest graphical demands.

Interestingly, the Ultra preset shows a different scaling behavior. At 1080p Ultra, the average is 136 FPS, which is almost identical to the High preset's 138 FPS. At 1440p Ultra, the system delivers 120 FPS, and at 4K Ultra, it drops to 82 FPS. The 4K Ultra result is the lowest of any tested configuration, and the gap between High and Ultra at 4K (86 vs 82 FPS) is just 4.7%. This compression in performance between settings at high resolution suggests that the GPU is finally being pressed into service, but the game engine itself may be imposing limits on how much work the Blackwell architecture can perform in this title.

The data implies that Thief is heavily single-thread bound in its core logic, which is typical for a game from its era. The i3-12100F's boost clock of 4.30 GHz provides solid single-core performance, but the four-core, eight-thread configuration means that any background tasks or engine threads that scale poorly will create stalls. The benchmark results show the processor's 3dmark_single_thread score of 893 and cinebench_r23_singlecore score of 1674, which are respectable but not exceptional. The real limitation appears when considering the cinebench_r15_multicore score of 1195, which is low compared to modern mid-range CPUs, indicating that multi-threaded workloads in the game engine may not be efficiently distributed.

Settings Recommendations

Examining the measured rows, the Medium preset emerges as the most balanced choice for this combination, particularly at higher resolutions. At 4K Medium, the system delivers 113 FPS average with a minimum of 96 FPS and a maximum of 130 FPS. This represents a 24.2% improvement over the Ultra preset's 82 FPS at the same resolution, while only sacrificing 31 FPS compared to Low's 125 FPS. The minimum frame rate of 96 FPS at 4K Medium is particularly noteworthy, as it suggests that even the worst-case dips remain comfortably above the 60 FPS threshold, providing a smooth experience without the need for adaptive sync.

At 1440p, the Medium preset again looks strong with 135 FPS average, 114 FPS minimum, and 155 FPS maximum. The gap between Medium and High at this resolution is 13 FPS (135 vs 122), which is a 10.7% penalty for the higher visual quality. The jump from Medium to Ultra at 1440p (120 FPS) costs another 11.1% performance. For players seeking the highest frame rates, the Low preset at 1440p produces 149 FPS, but the visual compromise is likely not worth the 14 FPS gain over Medium.

The 1080p results show that Medium (153 FPS) and High (138 FPS) are separated by 10.9%, while Ultra (136 FPS) is nearly identical to High. This suggests that at 1080p, the CPU is the limiting factor, as the GPU can handle the additional graphical load without significant frame rate changes. The minimum FPS at 1080p Medium is 130, which is higher than the average at 4K High (86 FPS), underscoring how resolution is the primary variable affecting performance in this configuration.

For a 4K display, the Medium preset is the clear recommendation. The 113 FPS average is well above the 82 FPS of Ultra, and the minimum of 96 FPS indicates that the system rarely stutters. If a player insists on maximum visual fidelity, the Ultra preset at 4K is still playable at 82 FPS, but the data shows that the extra graphical features consume significant GPU resources without proportional visual benefit in this older title.

GPU Role

The NVIDIA GeForce RTX 5090 is an absolute powerhouse on paper, with 21,760 shading units, 680 texture mapping units, and 176 ROPs. Its 104.8 TFLOPS FP32 performance and 1,636.8 GTexel/s texture rate are generational leaps over previous hardware. The 32 GB of GDDR7 memory on a 512-bit bus provides 1.79 TB/s of bandwidth, which is more than sufficient for any game released in the past decade, including Thief.

However, the benchmark data reveals that the RTX 5090 is not being pushed to its limits in this title. At 4K Ultra, the 82 FPS result, while respectable, does not reflect the GPU's theoretical capability. The passmark_g3d score of 39,650 and the 3dmark_steel_nomad_dx12 score of 18,355 indicate that the card can handle far more demanding workloads. The fact that Thief only uses a fraction of the VRAM—the game was released when 2-4 GB was standard—means that the 32 GB allocation is largely idle. The 1.79 TB/s bandwidth is also underutilized, as the game's texture streaming requirements are minimal by modern standards.

The GPU's clock behavior is interesting when considered against the frame rates. The base clock is 2017 MHz and the boost clock is 2407 MHz, but the actual sustained clocks during gameplay are not measured in the data. The fp32 performance of 104.8 TFLOPS suggests that the GPU should be able to render Thief at thousands of frames per second if the CPU could feed it data fast enough. Instead, the measured results show the GPU waiting on the processor, particularly at lower resolutions where the frame rate ceiling is set by the CPU's ability to issue draw calls.

The RTX 5090's 92.2 billion transistors on a 750 mm² die represent the pinnacle of Blackwell architecture, but in this context, it is like a sports car stuck in traffic. The 423.6 GPixel/s pixel rate and 104.8 TFLOPS compute are impressive specifications, yet the game engine's threading model appears unable to leverage this hardware efficiently. The GPU's passmark_directx_11 score of 341 and passmark_directx_12 score of 185 show that it excels in modern APIs, but Thief likely uses an older DirectX version that does not take full advantage of the card's features.

FAQ

Q: Why does the RTX 5090 not produce higher frame rates in Thief at 1080p?

A: The data shows that at 1080p, the frame rates across settings are nearly identical—169 FPS Low, 153 FPS Medium, 138 FPS High, and 136 FPS Ultra. This clustering indicates that the Intel Core i3-12100F, with its 4 cores and 8 threads, becomes the bottleneck before the GPU's capabilities are fully utilized. The CPU's cinebench_r20_singlecore score of 703 and geekbench_singlecore score of 1932 are modest, limiting the rate at which the game can submit rendering commands.

Q: Is the 4K Medium preset better than 4K High for this system?

A: Yes, the measured results show 113 FPS at 4K Medium versus 86 FPS at 4K High. The Medium preset also has a minimum FPS of 96, which is higher than the High preset's average. This suggests that the Medium settings reduce the GPU workload enough to allow the CPU to keep up, resulting in a smoother overall experience. The 31 FPS difference between Medium and Low at 4K is the largest gap in any resolution, indicating that the GPU begins to matter more at higher resolutions.

Q: How does the RTX 5090's 32 GB VRAM affect performance in this game?

A: The 32 GB of GDDR7 memory is vastly more than Thief requires, as the game was released in 2014 when 2-4 GB was common. The data does not show any performance difference attributable to VRAM capacity across the tested settings, suggesting that the game never approaches the memory limit. The 1.79 TB/s bandwidth provides ample headroom for texture streaming, but the game's asset sizes are small enough that this bandwidth is never a limiting factor.

Q: What is the best resolution to play Thief with this hardware?

A: Based on the data, 2560x1440 with Medium settings offers the best balance, delivering 135 FPS average with a 114 FPS minimum. This configuration provides a high frame rate that exceeds most display refresh rates while maintaining smoothness. At 4K, the Medium preset's 113 FPS is also excellent, but the 96 FPS minimum in the data suggests that occasional dips may be more noticeable on high-refresh 4K panels.

Q: Does the Ultra preset provide any benefit over High in this game?

A: The measured results show minimal differences between High and Ultra at most resolutions. At 1080p, High is 138 FPS and Ultra is 136 FPS, a negligible 1.5% difference. At 1440p, High is 122 FPS and Ultra is 120 FPS, also a 1.6% difference. The only notable gap is at 4K, where High achieves 86 FPS and Ultra drops to 82 FPS, a 4.7% penalty. This suggests that the Ultra preset's additional effects do not meaningfully improve visuals while consuming GPU resources.

Q: Why is the 4K Low preset slower than the 4K Medium preset in terms of relative scaling?

A: At 4K, Low produces 125 FPS while Medium produces 113 FPS, a difference of 12 FPS. In percentage terms, Low is 10.6% faster. However, this gap is smaller than the gap between 1080p Low (169 FPS) and 1080p Medium (153 FPS), which is also 12 FPS but represents only a 7.8% difference. This implies that at 4K, the GPU is becoming more of a limiting factor, and the additional graphics work in Medium does not reduce the frame rate as much as it might at lower resolutions where the CPU is the bottleneck.

How This Combo Ranks

The combination of the Intel Core i3-12100F and NVIDIA GeForce RTX 5090 ranks 467th out of 3,720 tested combinations in Thief. This places it in the top 12.5% of all configurations, which is a strong showing but not as high as one might expect given the GPU's 92nd percentile ranking among all GPUs. The CPU's 68th percentile ranking among all CPUs drags the overall combo down, highlighting the imbalance in this pairing.

The CPU's nearest rivals in the benchmark database include the Intel Core 3 N355 with an average score of 13,492 and a 0% delta, and the Intel Core i5-9500 with 13,452 and a 0.3% delta. The i3-12100F's average benchmark score of 13,494 is nearly identical to these processors, suggesting that its performance characteristics are well understood. The GPU's nearest rivals include the NVIDIA Tesla P100 PCIe 16 GB with an average score of 79,605 and a 0.3% delta, indicating that the RTX 5090's 79,842 average score is just above this enterprise-class card.

The rank of 467 out of 3,720 means that the combo outperforms 3,253 other combinations tested. This is an impressive result for a system with a $97 launch MSRP CPU, but the data shows that the RTX 5090 is not being fully utilized. If the CPU were upgraded to a higher-core-count model, the combo rank would likely improve significantly, as the frame rates at 1080p suggest that the GPU is waiting for the processor.

CPU Role

The Intel Core i3-12100F is a 4-core, 8-thread processor based on the Alder Lake architecture, manufactured on Intel's 10 nm process. Its base clock of 3.30 GHz and boost clock of 4.30 GHz provide solid single-thread performance, as evidenced by the cinebench_r23_singlecore score of 1,674 and the geekbench_singlecore score of 1,932. The 12 MB of shared L3 cache and 1.25 MB L2 cache per core are adequate for gaming workloads, but the processor's overall performance is limited by its core count.

In Thief, the CPU's role is critical because the game's engine, released in 2014, was designed for older multi-core processors. The 3dmark_4_threads score of 2,859 and 3dmark_8_threads score of 4,026 show that the processor scales reasonably well with additional threads, but the 3dmark_max_threads score of 4,023 indicates that there is no performance gain beyond 8 threads. This suggests that the game may not effectively utilize more than 4-6 threads, making the i3-12100F's 4 cores a potential limitation.

The cinebench_r20_multicore score of 4,981 and cinebench_r15_multicore score of 1,195 are modest compared to modern 6-core and 8-core processors. This becomes evident in the game's frame rates at lower resolutions, where the CPU is the bottleneck. At 1080p, the system cannot exceed 169 FPS even on Low settings, which is far below what the RTX 5090 could theoretically produce. The passmark_multithread score of 14,015 and passmark_single_thread score of 3,444 further confirm that the CPU's single-thread performance is its strongest asset, but multi-threaded workloads expose its limitations.

The processor's 58 W TDP and support for both DDR4 and DDR5 memory are notable, but the PCIe Gen 5 interface with 16 lanes ensures that the RTX 5090 has sufficient bandwidth for data transfer. The 3dmark_16_threads score of 4,051 shows that the CPU's performance does not improve beyond 8 threads, which is expected for a 4-core processor with Hyper-Threading. In the context of Thief, this means that the CPU is likely the limiting factor in most scenarios, particularly at 1080p and 1440p resolutions.

Measured FPS Breakdown

The complete dataset for this combination in Thief is as follows:

1920x1080 (1080p):

• Low: 169 FPS average
• Medium: 153 FPS average, 130 FPS minimum, 175 FPS maximum
• High: 138 FPS average
• Ultra: 136 FPS average

2560x1440 (1440p):

• Low: 149 FPS average
• Medium: 135 FPS average, 114 FPS minimum, 155 FPS maximum
• High: 122 FPS average
• Ultra: 120 FPS average

3840x2160 (4K):

• Low: 125 FPS average
• Medium: 113 FPS average, 96 FPS minimum, 130 FPS maximum
• High: 86 FPS average
• Ultra: 82 FPS average

The 1080p results show a tight grouping between High and Ultra, with only 2 FPS separating them. The Medium preset at 1080p is 15 FPS faster than High, while Low is 16 FPS faster than Medium. This pattern suggests that the CPU is the limiting factor, as the GPU can handle the graphical load without impacting frame rates significantly.

At 1440p, the differences between settings become more pronounced. Low is 14 FPS faster than Medium, and Medium is 13 FPS faster than High. The gap between High and Ultra narrows to just 2 FPS, again indicating CPU limitation. The minimum frame rate at 1440p Medium (114 FPS) is higher than the average at 1080p Ultra (136 FPS), showing that resolution has a more significant impact than settings at this level.

The 4K results reveal the GPU starting to assert its role. Low is 12 FPS faster than Medium, and Medium is 27 FPS faster than High. The gap between High and Ultra at 4K is 4 FPS, which is the largest difference between these two settings at any resolution. This indicates that at 4K, the GPU is finally becoming the bottleneck, and the additional graphical features in Ultra have a measurable impact. The minimum FPS at 4K Medium (96) is still well above playable thresholds, making this the recommended configuration for 4K displays.